WattWise：開源本機 AI 電費核查系統

Hacktoberfest Open-Source AI Challenge: Ghost in the Machine Submission 👻🌼

This is a submission for the Hacktoberfest Open-Source AI Challenge Week 4: Ghost in the Machine

What I Built

WattWise is an open-source, locally-hosted AI verification system that acts as your personal electricity bill auditor.

By leveraging an original Model Harness and strict Agent Skills, WattWise allows consumers to securely audit their electricity bills for mathematical errors, impossible consumption readings, and tariff mistakes—all without sacrificing their personal data privacy.

Demo

Code

https://github.com/DevNomad26/wattwise-harness

How I Built It

The Model Harness
Small language models (like 4B parameters) are incredibly fast and fit on consumer hardware, but they suffer from hallucinations and struggle with floating-point math. To solve this, we built an original, multi-turn conversational ReAct Agent Harness. Our harness intercepts hallucinations, maintains a persistent session history across turns, and acts as the strict orchestrator between the LLM and the real world.
Agent Skills
Instead of relying on prompt engineering, the agent dynamically loads custom Skills formatted to a standard SKILL.md structure. This enables complex, multi-step workflows such as:

bill-checker: Validates physically impossible consumption rates and billing discrepancies.
complaint-letter: Drafts automated formal dispute and refund notices to DISCOMs.
solar-roi-calculator: Estimates rooftop solar capacity (kW), generation, and payback periods.
energy-saving-advisor: Recommends appliance-level actions to reduce usage and drop to cheaper tariff slabs.
explain-bill: Breaks down fixed charges, energy slabs, and duties in clear, plain language.

Why Does Open Innovation Matter?

Electricity bills are notoriously difficult to decipher, and mistakes by distribution companies (DISCOMs) are common. Consumers often overpay because they cannot mathematically verify complex tariff structures, slab rates, or net-metering (solar) adjustments.

Because electricity bills contain sensitive personal addresses, account numbers, and financial data, uploading them to third-party cloud APIs (like OpenAI or Anthropic) is a privacy risk.

WattWise solves this by running 100% locally on your machine. We use open-weight models (qwen3.5:4b) running on Ollama, ensuring your data never leaves your computer.

Prize Categories

GitHub
Qwen3.5

Team Members :
Udayan Amipara(me)
Akash Wadhvani
Naman Sawnani